An efficient way to compute Feynman integrals is to derive integration-by-parts relations, choose a basis of master integrals, and put their differential equations in canonical form. In the polylogarithmic case, this is closely tied to the existence of integrals with exclusively logarithmic differential forms and constant leading singularities. These notes describe how this picture extends beyond multiple polylogarithms. The search for canonical integrals can be phrased as an analysis of the differential forms that appear in the relevant integrands. For elliptic curves and more general geometries, new transcendental normalizations appear, which we show are related to the leading singularities of the corresponding Feynman integrals.